Heat transfers at work

This trial studies the effect of heat-indexed anticipatory and day-of cash transfers and early heat warnings on informal workers largely employed in small factories in Delhi NCR. Cash transfers and early warnings are triggered when daily temperatures are forecasted to reach or exceed 40°C. Payments amount to ₹500 per trigger day – a little more than the average daily wage in our sample – with at most 3 payouts per respondent per week and a maximum of 6 payouts per season.

2026-06-06

2026-07-31

Labour supply and health

Our key outcome variables relate to the labor supply and health of workers. Specifically, we will look at days and hours worked, and heat-induced illness.

We plan to collect these outcomes through two types of data:

1. Daily time-use survey, completed either over text or by phone
2. Recall-based measures collected during an in-person endline

Our secondary outcomes include adoption of heat-adaptive behaviors (e.g., purchasing Oral Rehydration Solutions and air coolers) and indicators of financial distress (e.g., borrowing to cover daily expenses).

We implement a two-stage, four-arm cluster randomised experiment across 10 informal settlements in Delhi NCR, India. Clusters of co-workers are randomly assigned to one of four arms:

Arm 1: Pure control group. No intervention. Respondents are surveyed but receive no information or transfers.
Arm 2: Information only: Receives heat forecast alerts (early warning) at the start of the week but no cash transfers.
Arm 3: Anticipatory cash: Receives ₹500 per forecasted trigger day at the start of each week (up to three payments), accompanied by heat forecast alerts.
Arm 4: Day-of cash: Receives ₹500 on each trigger day (up to three per week), plus an evening notification confirming the following day's forecast will exceed 40°C and that payment has been triggered.

Arms 2-4 each have a 50% and 100% saturation variant. In 100%-saturation clusters, every sampled worker within a given cluster is offered the intervention. In 50%-saturation clusters, only half are randomly selected for the intervention. This design allows estimation of within-workplace spillover effects.

The randomization is stratified by neighbourhood location.

The design allows for four key comparisons:

1. Arm 1 versus Arm 2 isolates the causal effect of heat information alone;
2. Arm 2 versus Arm 3 identifies the value of anticipatory cash over information alone;
3. Arm 3 versus Arm 4 identifies the value of transfer timing (anticipatory versus day-of), noting that anticipatory cash is paid as a lump sum at the start of the week, while day-of cash may be distributed across multiple days.

We estimate the intention-to-treat (ITT) effects of treatment assignment using the specification below. Let arm 1 (pure control) be the omitted category.

Y_ijts = α + β1×Arm2_i + β2×Arm3_i + β3×Arm4_i + λ_s + ßX_ijt + ε_ijts

Where:
Y_ijts: outcome of interest for individual i at workplace j, living in neighbourhood s, measured in period t.
Parameters β1, β2, and β3 capture the average ITT effects of being assigned to the information-only, anticipatory cash, and day-of cash arms respectively, relative to the pure control group.
X_ijt is a vector of control variables.
λ_s are neighbourhood fixed effects.
Standard errors are clustered at workplace level.

We pre-specify the following dimensions of heterogeneity by baseline:
1. Heat exposure at work
2. Gender
3. Earnings

Not available

Randomisation done in office by a computer, using the listing survey as the sampling frame.

Workplace

Yes

800-1000 clusters with an average of 2-3 respondents per cluster

Approximately 1,800-2,400 respondents

Approximately 450-600 per arm across four arms (pure control, information only, anticipatory cash, day-of cash). Approximately 350-450 respondents per informal settlement.